Anthropic just released Claude Opus 4.8 today and honestly, it came out of nowhere well, sort of. It’s only been about six weeks since Opus 4.7 dropped in April, so the quick turnaround alone tells you something about how aggressively the company is moving right now.
I’ve gone through everything Anthropic shared about this release and there are a few things here that genuinely stand out from the usual we improved performance press release language. Let me break it down.
So What Actually Changed in Claude Opus 4.8?
The short answer: coding got better, it runs faster and cheaper in fast mode and you now have more control over how hard the model think before answering you.
The longer answer is more interesting.
On the benchmark side, Opus 4.8 scored a 69.2% agentic coding score — up from 64.3% on Opus 4.7. That’s not a tiny tweak. For developers using Claude to write, debug, or review code, that gap is noticeable in day-to-day use. Knowledge work performance also jumped from 1,753 to 1,890 on Anthropic’s internal scoring. A meaningful leap for things like research, drafting and analysis tasks.
Fast Mode Got a Serious Upgrade
Here’s the part that’ll matter to most users: fast mode is now 2.5x quicker and costs three times less than the previous fast version. Standard pricing stays the same — $5 per million input tokens, $25 per million output tokens. But if you were hesitant about fast mode before because of the cost, that barrier is mostly gone now.
This is Anthropic directly responding to user feedback. Speed and price have been the two loudest complaints about premium AI models, and they tackled both in one shot.
You Can Now Tell Claude How Hard to Think
This is genuinely new behavior. Anthropic added effort controls — a slider-style setting next to the model selector in Claude.ai that lets you pick how much reasoning effort goes into a response:
- High effort is the default for most tasks
- Extra effort kicks in for complex research or multi-step problems
- Max effort throws everything at your hardest questions
On the surface it sounds like a small UI addition, but it’s actually a pretty smart move. You’re no longer paying for max-effort thinking on a simple question and you’re not getting a lazy response when you need Claude to really dig in. You decide.
Dynamic Workflows in Claude Code
If you’re a developer, pay attention to this one. Anthropic introduced dynamic workflows in Claude Code — still in research preview, but already impressive. The idea is that Claude can now plan out a large project independently, spin up hundreds of parallel sub-agents inside a single session and verify results before it reports back to you.
What that means practically: you can hand off a genuinely complex engineering task — something that would normally take hours of back-and-forth and Claude handles the orchestration itself. It figures out what needs to be done in what order, runs multiple threads simultaneously and checks its own work. That’s a different level of autonomy compared to what most AI tools offer right now.
Context Window and Where You Can Use It
Opus 4.8 supports a 1 million token context window on the Claude API, Amazon Bedrock and Google Vertex AI — perfect for processing large documents, entire codebases, or lengthy research materials in one shot. Microsoft Foundry users get 200k tokens.
It’s also live today on GitHub Copilot for Pro+, Business and Enterprise users. Which means a huge chunk of developers already have access without needing a separate subscription.
The Bigger Picture You Should Know
Anthropic is releasing major model updates roughly every six weeks right now. That pace is deliberate. The company recently closed a funding round that pushed its valuation to $900 billion — ahead of OpenAI’s $730 billion and they’re spending that capital on exactly this: faster iteration, better models, lower prices.
They also confirmed a Mythos cybersecurity model is coming to all customers in the coming weeks. It’s been in limited access since early April, and broader availability will open up some seriously advanced threat analysis capabilities for enterprise teams.
This isn’t a company coasting. Every release right now is pushing something forward in a concrete way.
Who’s This For?
Look if you’re a developer, Opus 4.8 is probably the best coding-focused AI model you can use right now. The benchmark jumps are real, and dynamic workflows in Claude Code is the kind of feature that saves hours, not minutes.
If you’re a researcher, analyst or someone doing heavy knowledge work, the improved reasoning scores and the honest-uncertainty behavior (Claude now flags when it’s unsure rather than guessing confidently) make it more trustworthy for serious tasks.
And if you’re just a power user who wants more control over how your AI assistant behaves, effort controls are worth experimenting with. It’s a small change that actually adds up when you’re using the tool all day.
Quick FAQ
Is Claude Opus 4.8 available right now?
Yes — it went live globally on May 28, 2026 across Claude.ai, the API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry and GitHub Copilot.
What does Claude Opus 4.8 cost?
Standard: $5/million input tokens, $25/million output tokens. Fast mode: $10 input / $50 output per million tokens — but that’s 3x cheaper than old fast mode pricing.
How different is it from Opus 4.7?
Meaningfully different, especially in coding, knowledge work and autonomous task handling. The effort controls and dynamic workflows are genuinely new behavior, not just renamed features.
What’s the context window for Claude Opus 4.8?
1 million tokens on most platforms (Claude API, Bedrock, Vertex AI). Microsoft Foundry caps at 200k.
